Proof that Elon while a genius in many technical topics he can be poor at general knowledge. EVERYONE knows that to burry stock news (or any other news) release it on a Friday afternoon. He should have waited until Monday afternoon but like a petulant child he couldn't wait.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
If this product reveal like the Roadster, the Semi or the CT or is it like Autonomy day 3?
@willow_hiller? @uscbucsfan? Reuters? Anyone?
@willow_hiller? @uscbucsfan? Reuters? Anyone?
AlanSubie4Life
Efficiency Obsessed Member
He’s going to blame Reuters, a dying news organization. Only Twitter is a worthy substitute.
diplomat33
Average guy who loves autonomous vehicles
How come Tesla needs to unveil a dedicated robotaxi when all our existing cars were supposed to be robotaxis? 
I was becoming concerned that vision-only self driving would have obvious technical driving failures. The FSD 11.x looked promising, but could go either way. FSD 12.3.3 is very good at routine driving. It looks like it now could succeed without radar or lidar for primary driving.
The redundancy issue is another story. That could be the actual limits of the Tesla approach that prevent regulatory acceptance for a driverless vehicle.
With FSD 12.3.3 I find myself too easily distracted while in the driver seat. The car just drives too well. But occasionally it makes a terrible error which requires immediate human control. Even so, I find that I’m using FSD with navigation more than driving myself. Now I berate myself when I “forget” to enter the location I want to go to. There are a lot of straight roads here, and I’ve gotten caught thinking that I’m going straight thru a roundabout without navigation active and the car basically makes a right turn. Oops.
Now the question is how good can vision-only become in finite time (one - two years). The answer to whether the vision-only technology can be level three FSD is what is the limit of the neural net on Tesla HW3/4? Now I think it’s within reach. The car is already better than the average driver in nearly all cases. Altho the average driver is better at parking in a house garage.
So here is my list of missing features/issues ( features where a human is routinely better):
* merging onto a freeway with traffic
* parking in a garage
* departing from a garage parking spot, down a private driveway to a public road
* emergency vehicles on or near the road
* parking at final destination
* badly behaved traffic ala LA/NYC/Boston
* speed control near speed traps and just in general
* avoiding potholes and debris on the road
* avoiding small animals on the road
I’m sure there are more, but I find it’s now my preferred driving method in decent weather, and seems like the above issues are technically resolvable with current hardware.
The future looks interesting.
The redundancy issue is another story. That could be the actual limits of the Tesla approach that prevent regulatory acceptance for a driverless vehicle.
With FSD 12.3.3 I find myself too easily distracted while in the driver seat. The car just drives too well. But occasionally it makes a terrible error which requires immediate human control. Even so, I find that I’m using FSD with navigation more than driving myself. Now I berate myself when I “forget” to enter the location I want to go to. There are a lot of straight roads here, and I’ve gotten caught thinking that I’m going straight thru a roundabout without navigation active and the car basically makes a right turn. Oops.
Now the question is how good can vision-only become in finite time (one - two years). The answer to whether the vision-only technology can be level three FSD is what is the limit of the neural net on Tesla HW3/4? Now I think it’s within reach. The car is already better than the average driver in nearly all cases. Altho the average driver is better at parking in a house garage.
So here is my list of missing features/issues ( features where a human is routinely better):
* merging onto a freeway with traffic
* parking in a garage
* departing from a garage parking spot, down a private driveway to a public road
* emergency vehicles on or near the road
* parking at final destination
* badly behaved traffic ala LA/NYC/Boston
* speed control near speed traps and just in general
* avoiding potholes and debris on the road
* avoiding small animals on the road
I’m sure there are more, but I find it’s now my preferred driving method in decent weather, and seems like the above issues are technically resolvable with current hardware.
The future looks interesting.
Doggydogworld
Active Member
I believe he'll demo a podcar driving in traffic on public streets without a safety driver. Not in California, but a state with lax regulations like NV or AZ.
At this point a static podcar display combined with more staged videos and a rehash of five year old promises from Autonomy Day 2019 just won't cut it. Has to be a live demo in traffic to move the needle. Maybe a closed course thing instead of public streets, but even that is weak sauce.
diplomat33
Average guy who loves autonomous vehicles
willow_hiller
Well-Known Member
Hardly hard truths when he sets up a false straw man in the 4th sentence.
"Teslas are failing at basic driving, not edge cases... Parked Firetrucks, cement barriers"
For the parked firetrucks, I think he's referencing a Bay Area crash that occurred in 2023 on Autopilot. Not sure where he's getting "cement barriers." But either way, the man doesn't even seem to know the difference between Autopilot and FSD.
AlanSubie4Life
Efficiency Obsessed Member
To be fair, this IS true. It's really, really good. But it routinely fails requiring safety disengagements.
Yes, dumb to reference old problems. Plenty of new problems!
It was in his thread.
Doggydogworld
Active Member
Yes, with chase car and/or passengers in car with big red stop button. Yandex did empty driver's seat / big red button demos in Vegas years ago. No lengthy permit process required. Keep the speeds low so any accident is minor. Throw out lots of AI jargon to impress the gullible. Announce Robotaxi service in 50 cities by 2026 vs. Waymo's doomed approach that only allows them to add one city every year or three.
Great analogy. I think Tesla will get to waymo capabilities before a 9 year gap. Now they do have some abilities Waymo doesn't, the building cars.
Probably talking about a car without steering wheels or pedals. Under current rules only 2500 units is allowed to be exempt per manufacturer for such vehicles, so it pretty much has to be a separate dedicated vehicle.How come Tesla needs to unveil a dedicated robotaxi when all our existing cars were supposed to be robotaxis?
https://www.reuters.com/business/au...-self-driving-car-deployment-plan-2023-07-12/
Cruise's application didn't seem to be approved (especially given the turn of events), but the articles referenced to the same 2500 unit number.
https://www.reuters.com/business/autos-transportation/investors-pull-away-gms-cruise-bet-2023-11-10/
Oh No Oh No He saw the car headed to the Wall and didn't try to Stop it.
diplomat33
Average guy who loves autonomous vehicles
Final video of highway testing from Ghost Autonomy before they shut down.
I post because I think it is informative. We can see the debug info on the screen. We can see lane confidence was only ~80-85% and it looks like object permanence was pretty bad too. We also see object velocity would come and go on the screen. And the safety driver disengages for highway transitions.
Not surprised that they shut down considering that their L2 highway was far behind. Even if they did get their L2 highway to be good enough, it is unlikely they would find many OEMs to license it. Ghost could have tried to sell a kit direct to customers but I think the market is very crowded now. The business model would be tough.
I post because I think it is informative. We can see the debug info on the screen. We can see lane confidence was only ~80-85% and it looks like object permanence was pretty bad too. We also see object velocity would come and go on the screen. And the safety driver disengages for highway transitions.
Not surprised that they shut down considering that their L2 highway was far behind. Even if they did get their L2 highway to be good enough, it is unlikely they would find many OEMs to license it. Ghost could have tried to sell a kit direct to customers but I think the market is very crowded now. The business model would be tough.
Daniel in SD
(supervised)
Those both say Cruise requested approval for 2500 vehicles not that there’s some sort of limit.
If you need another article that explicitly pointed it out, here's one:
"NHTSA only grants 2,500 such exemptions each year, but there is legislation to increase that number to 25,000."
Cruise nears approval to mass-produce robotaxis with no steering wheel, pedals | TechCrunch
If you need a direct NHTSA source, here's a press release from earlier applications:
"There is an exemption cap of 2,500 vehicles per manufacturer."
U.S. Department of Transportation Seeks Public Comment on GM and Nuro Automated Vehicle Petitions | NHTSA
Similar threads
- Article
